It seems like any product offering public ratings will end up having fake ratings.
What have you considered to make sure reviews remain legit?
Some of the early thinking I've done is around the system being able to verify that a user did in fact apply to a company (assuming company is a customer of Rolepad). That provides some initial guarantees about the author of the data. With some ATS integration, there could be fairly concrete numbers on "how quickly did X happen" that don't even need to be sourced from the candidate. That said, there is always a bit of a "he said, she said" with reviews, and I haven't fully figured out an obvious way forward here. I think the general sentiment I've seen is that Glassdoor and Yelp reviews should be treated with a grain of salt, but they're still better than nothing.
My employer keeps asking us to "leave honest reviews" to help with hiring efforts. In practice that means they want us to leave GOOD reviews to raise rating scores on websites such as Glassdoor.
Sorry HR, I'm not going to do your work for you.
We eventually got it removed, but it took an extreme amount of effort to convince them to take it down.
I can't imagine how difficult it would be to take a review down that isn't as easy to prove illegitimate.
This was in 2020, so maybe things have devolved since then..
I was doing business studies research stuff when blockchain was all the rage. The accounting integrity was a major buzzword around that circle. Nobody came up with anything concrete or utilitarian.
The challenge is that you need to remove malicious reviews or spams. But at the same time that review system cannot be exploited.
Is there a systematic way to do this?
Throwing an idea here. What if there was a distributable append only database that obfuscated the content (reviews, company, user, company etc) with hash function only showed the action (new review, update review, delete review etc.)?
The database should be shared across multiple stakeholders as single source database is easily corruptible. For each actions to go through, these stakeholder databases need to be verify them. Tis verification process essentially links these databases to a single system. Then once the verification is done the review can be added.
To incentivize the storing and verification we need to offer an incentive of sort that is not money. As money corrupts everything and is the source of all evil it should be a form of token of gratitude that can be essentially transacted with other participants a form of pseudo currency. The value of the pseudo currency derives from everyone believing in the system.
In fact we can probably create an algorithm that automates the verification process using LLM algorithm and historic patterns. Those LLM algorithms can run on the stakeholder machines preferably on GPU.
Would a system like this work?